pith:WFMWIHAP
Can RL Teach Long-Horizon Reasoning to LLMs? Expressiveness Is Key
Reinforcement learning overcomes LLM long-horizon reasoning limits when training uses more expressive logic.
arxiv:2605.06638 v3 · 2026-05-07 · cs.AI · cs.CL
Add to your LaTeX paper
\usepackage{pith}
\pithnumber{WFMWIHAPMG25JCPAX3GTQLQBCB}
Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge
Record completeness
Claims
LLM shortcomings in long-horizon reasoning are not fundamental to the underlying architecture, and can be addressed by improved training methodology and data.
That performance on the synthetic ScaleLogic tasks and their transfer to downstream benchmarks is a faithful proxy for the long-horizon reasoning difficulties encountered in real-world applications.
RL training compute for logical reasoning follows a power law in proof depth whose exponent rises with logic expressiveness, and more expressive training yields larger gains on downstream benchmarks.
Formal links
Receipt and verification
| First computed | 2026-05-20T00:03:14.548748Z |
|---|---|
| Builder | pith-number-builder-2026-05-17-v1 |
| Signature | Pith Ed25519
(pith-v1-2026-05) · public key |
| Schema | pith-number/v1.0 |
Canonical hash
b159641c0f61b5d489e0becd382e0110545ada71a70564aa4c8e21446ba06469
Aliases
· · · · ·Agent API
Verify this Pith Number yourself
curl -sH 'Accept: application/ld+json' https://pith.science/pith/WFMWIHAPMG25JCPAX3GTQLQBCB \
| jq -c '.canonical_record' \
| python3 -c "import sys,json,hashlib; b=json.dumps(json.loads(sys.stdin.read()), sort_keys=True, separators=(',',':'), ensure_ascii=False).encode(); print(hashlib.sha256(b).hexdigest())"
# expect: b159641c0f61b5d489e0becd382e0110545ada71a70564aa4c8e21446ba06469
Canonical record JSON
{
"metadata": {
"abstract_canon_sha256": "cdfb6df287821caf8a26addbb8a2c859e647967bb784ed52a71031bd1bf95e5b",
"cross_cats_sorted": [
"cs.CL"
],
"license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
"primary_cat": "cs.AI",
"submitted_at": "2026-05-07T17:48:42Z",
"title_canon_sha256": "cadbde38064688eda45f9f9931dbc3830cc800c0b2b88f2672c696b2c5e12ff0"
},
"schema_version": "1.0",
"source": {
"id": "2605.06638",
"kind": "arxiv",
"version": 3
}
}